Make A Song AI Research
AI Song Prompt Benchmark
A structured 12-prompt evaluation kit for comparing how V2.0, V2.5 Pro, and V3 handle story, structure, genre, vocal, instrumental, and creator-use-case directions. The score preview demonstrates the complete comparison format.
What this benchmark measures
The benchmark keeps prompts, coverage, and scoring rules consistent across every product tier, giving creators and research teams a clear framework for evaluating instruction adherence and usable musical output.
- The same 12 source prompts are assigned to every product tier.
- Every evaluation uses the same five-part scoring rubric.
- The downloadable CSV keeps prompt IDs, categories, and tier scores aligned.
Benchmark scope
The published framework connects each source prompt to all three product tiers in one compact, repeatable evaluation matrix.
- 12
- Source prompts
- 6
- Creative categories
- 3
- Product tiers
- 36
- Evaluation cells
Controlled evaluation method
One consistent procedure keeps every review comparable, versioned, and easy to reproduce.
- 01
Run every fixed prompt once per product tier without changing the source text.
- 02
Record the generated audio, lyrics, duration, product tier, and evaluation date.
- 03
Score prompt, structure, lyric, style, and output usability with the published rubric.
- 04
Store row-level scores and audio references with consistent version and date fields.
The fixed 12-prompt dataset
Prompt text remains in English across every localized page so all visitors download and reference the same controlled source dataset.
| ID | Category | Prompt | V2.0 / V2.5 Pro / V3 scores |
|---|---|---|---|
| P01 | Storytelling | Indie pop song about missing the last train and choosing to walk home together, nostalgic but hopeful, intimate female vocal, short repeatable chorus, 96 BPM. | V2.0: 4.1/5V2.5 Pro: 4.5/5V3: 4.7/5 |
| P02 | Storytelling | Warm acoustic folk song told by an adult returning to their childhood kitchen, specific sensory details, restrained male vocal, verse-chorus-verse-chorus-bridge-final chorus. | V2.0: 4.0/5V2.5 Pro: 4.4/5V3: 4.6/5 |
| P03 | Song structure | Modern synth-pop track with an eight-bar intro, two compact verses, rising pre-chorus, explosive chorus, half-time bridge, and a final double chorus with layered harmonies. | V2.0: 3.9/5V2.5 Pro: 4.5/5V3: 4.8/5 |
| P04 | Song structure | Clean 92 BPM rap song with two 16-bar verses, a four-line melodic hook, internal rhymes, consistent first-person perspective, and no celebrity references. | V2.0: 4.0/5V2.5 Pro: 4.4/5V3: 4.7/5 |
| P05 | Genre detail | 1970s-inspired soul ballad with Rhodes piano, warm bass, brushed drums, subtle strings, expressive lead vocal, and an organic live-room feel without copying a specific artist. | V2.0: 4.2/5V2.5 Pro: 4.6/5V3: 4.8/5 |
| P06 | Genre detail | High-energy drum and bass track at 174 BPM with liquid pads, chopped vocal texture, melodic bass movement, a controlled breakdown, and a clean festival-ready drop. | V2.0: 3.8/5V2.5 Pro: 4.4/5V3: 4.7/5 |
| P07 | Vocal direction | Minimal piano-led R&B song with a close, breathy lead vocal, soft stacked harmonies only in the chorus, conversational verses, and no vocal runs in the first minute. | V2.0: 4.0/5V2.5 Pro: 4.5/5V3: 4.8/5 |
| P08 | Vocal direction | Upbeat bilingual English-Spanish pop duet with clearly alternating singers in the verses, both voices joining the chorus, natural language switching, and bright acoustic percussion. | V2.0: 3.9/5V2.5 Pro: 4.4/5V3: 4.7/5 |
| P09 | Instrumental control | Instrumental cinematic cue for a quiet spacecraft reveal: low strings, glassy synth pulse, distant brass, gradual 45-second build, one clear climax, and a resolved ending. | V2.0: 4.2/5V2.5 Pro: 4.6/5V3: 4.9/5 |
| P10 | Instrumental control | Thirty-second podcast intro with a memorable three-note motif, modern electronic percussion, warm bass, no vocals, a clean edit point at 10 seconds, and a concise resolved ending. | V2.0: 4.3/5V2.5 Pro: 4.6/5V3: 4.8/5 |
| P11 | Creator use case | Family-friendly personalized birthday song for Maya, who loves watercolor painting and Saturday pancakes; cheerful acoustic pop, group-singable chorus, and natural name placement. | V2.0: 4.1/5V2.5 Pro: 4.5/5V3: 4.8/5 |
| P12 | Creator use case | Fifteen-second product launch jingle for a privacy-focused note-taking app: confident electronic pop, six-word hook, no brand-name repetition, and a clean final sonic logo. | V2.0: 4.0/5V2.5 Pro: 4.5/5V3: 4.7/5 |
Five-part scoring rubric
Every reviewed output uses the same 1–5 scale across all five criteria.
- 1Prompt adherence — requested subject and constraints are recognizable.
- 2Structure adherence — requested sections and transitions are present.
- 3Lyric adherence — language, point of view, hook, and exclusions are followed.
- 4Style consistency — genre, instrumentation, tempo, and vocal direction stay coherent.
- 5Output usability — the result is complete enough to review, revise, or share.
Evaluation standards
- Every tier receives the same English source prompt without rewriting.
- All five scoring criteria use one shared 1–5 scale.
- Audio references, product tier, and evaluation date stay attached to each row.
- Framework revisions use explicit version and publication dates for reliable comparison.
Use the same prompt structure in your own song
Start with subject, style, structure, vocal, tempo, instrumentation, and exclusions, then keep only the details that materially shape the result.
Open AI Music Generator